Keep the model. Change the ownership.

Rain meets a city, discovers several million places it is not meant to be, and becomes an engineering problem. swmmrs translates the EPA SWMM solver into Rust. The hydraulic model stays familiar. Each simulation gets its own state, and the compiler stands by the airlock asking who owns what.

Hydraulic universe Near-identical results against EPA SWMM

Engine room Faster engine in the published benchmarks

Flight controls Python, CLI, and browser with isolated simulations

Launch from terminal Pilot it with Python Build with JS Open local workspace

Evidence, not prophecy

EPA-faithful results, faster runs, and APIs meant for this century.

swmmrs produces near-identical results to EPA SWMM. In the published real-world benchmarks, it also runs faster. Native Python APIs, isolated concurrent simulations, interactive controls, checkpoints, and typed results make the solver usable outside a single command-line voyage.

The regression evidence and real-world benchmark results contain the receipts.

These claims apply to the published engine versions, workloads, hardware, and comparison method. No finite suite visits every possible timeline. Compare critical production results against a trusted EPA SWMM baseline.

One prime directive

Parity by default. Hydraulic innovation by explicit opt-in.

The default path follows EPA SWMM. New hydraulic ideas can exist, but they live in clearly marked alternate timelines that the user must choose.

Default EPA path	Opt-in experiments
Existing EPA SWMM models should run with minimal result changes.	New hydraulic formulations and algorithms may join the project.
Calculations, evaluation order, edge cases, and input semantics stay intact.	The user must enable every alternative explicitly.
Differences remain reviewable beside the upstream C source.	An experiment never silently replaces a default. Its behavior, validation, and tradeoffs must be documented.

Parity is the baseline, not the ceiling. A model uses non-EPA hydraulics only when its user asks for them. No wormholes, no secret rerouting of the timeline.

Why put SWMM in Rust?

I began swmmrs to learn Rust on software I already use and respect. A to-do list would have involved fewer differential equations, but it would also have been a to-do list.

Rust makes each simulation own its files, objects, lifecycle, statistics, and mutable solver state. One simulation no longer shares a giant control panel with every other simulation in the process.

The translation keeps function order, names, comments, domain language, and floating-point order recognizable beside EPA SWMM. That restraint is intentional. Reviewers can compare the two implementations without decoding a clever new civilization first.

The ownership work also gives the project typed APIs, isolated tests, and independent concurrent runs. Those are useful consequences, not an excuse to change the model.

What changed under the hull

Area	Improvement
State	Each `SwmmSimulation` owns one private `SwmmState` instead of process-global project state.
Python	Typed lifecycle, object collections, validated setters, runtime forcings, hotstarts, snapshots, and statistics.
Results	Live scalar reads for control, coherent snapshots for network sampling, and immutable cumulative records.
Concurrency	Independent simulations can run concurrently without sharing hydraulic state.
Extension policy	Non-EPA formulations may be developed, but remain strictly opt-in.

Regression validation

Matching one model proves approximately one model. So swmm-bench runs an end-to-end suite against EPA SWMM and compares parsed report tables with binary-output time series. The suite covers hydrology, hydraulics, controls, routing, water quality, and interface-file workflows.

The EPA SWMM coverage report shows that the suite exercises more than 87% of upstream solver lines and 67% of branches. Read how to interpret the validation or inspect the latest HTML regression report.

Real-world benchmarks

Accuracy matters, but so does getting the answer back before the deadline. The companion swmm-bench benchmark runs SWMM-compatible engines on large hydraulic models drawn from day-to-day modelling work. It reports simulation duration beside report and binary-output similarity.

On the tested workloads, swmmrs is faster than EPA SWMM while maintaining a small result distance from the original engine.

Read the benchmark scope and interpretation or inspect the latest HTML benchmark report. No galactic standards committee has issued a universal speed ranking. The findings belong to the engine versions, workloads, hardware, and run conditions in that report.

Known limits, not plot twists

Dynamic Wave is the largest structural departure from upstream C and remains the highest-priority parity area.
During routing, collect live values and snapshots. After routing stops, use OutputReader for report-period output.
See compatibility and limitations before production evaluation.

Choose an interface

Python

Pause time, inspect a node, change a control, or run several scenarios without making them share a brain.

Python docs →
JavaScript / TypeScript

Put the solver in a worker, keep it off the UI thread, and retain typed collections, controls, snapshots, and output downloads.

JavaScript / TypeScript docs →
Command line

Give the runner one input file. It returns a text report and SWMM binary output with very little ceremony.

CLI guide →
Concepts

Read how isolated state, post-open declarations, lifecycle, and continuation affect programmatic modeling before inventing a paradox.

Concepts overview →
Solver differences

See where state isolation, performance work, checkpoints, and typed interfaces differ from EPA SWMM while the default hydraulics stay put.

Compare with EPA SWMM →

Concepts and contributions

Start with Concepts for a modeller-focused explanation of programmatic state, post-open declarations, solver lifecycle, and continuation. When you are ready to send something back across the subspace relay, see Contributing.

Upstream projects

There was no alien archive behind swmmrs, only decades of work by the EPA SWMM developers and the Open Water Analytics community. Keeping the translation recognizable acknowledges that work and helps with parity review. The repository's third-party notices reproduce the OWA MIT license, the EPA public-domain statement, and contributor attribution.